A Novel System for Precise Grading of Glioma

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Gliomas are primary brain tumors with high mortality.
• Accurate glioma grading is critical for effective treatment planning.
• Current grading methods can be invasive or subjective.

Purpose of the Study:

• To develop a non-invasive, multimodal magnetic resonance (MR)-based computer-aided diagnostic (CAD) system for precise glioma grading.
• To differentiate between glioma Grades I, II, III, and IV.
• To establish a novel tool for objective glioma characterization.

Main Methods:

• Utilized three MR imaging modalities: contrast-enhanced T1-MR, T2-MR (FLAIR), and diffusion-weighted (DW-MR).
• Extracted morphological features (HOG, volume), textural features (GLRLM, GLCM), and functional features (ADC, contrast-enhancement slope).
• Employed a Gini impurity-based selection approach and a multi-layer perceptron artificial neural network (MLP-ANN) classifier.

Main Results:

• The developed glioma grading CAD (GG-CAD) system achieved 0.96 ± 0.02 quadratic-weighted Cohen's kappa and 95.8% ± 1.9% overall diagnostic accuracy using leave-one-subject-out (LOSO) cross-validation.
• Demonstrated outstanding diagnostic performance with k-fold stratified cross-validation (k=5 and 10).
• Outperformed alternative classifiers like Random Forests (RFs) and Support Vector Machines (SVMlin).

Conclusions:

• The proposed GG-CAD system is a feasible and effective non-invasive tool for objective glioma grading.
• The system's comprehensive feature extraction and selection contribute to its high diagnostic performance.
• This developed CAD system holds promise for precise, non-invasive glioma grading in clinical practice.